jungsin3 commited on
Commit
aa44345
·
verified ·
1 Parent(s): 62cd18f

Add model weights and configs

Browse files
chat_template.jinja ADDED
@@ -0,0 +1,162 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {#- KORMo-VL XML 도구판 chat template (2026-10-28 imsi-docker2-0, 사용자 10-27: 1.5m 부터 XML 방식 · "잘 모르면 qwen3.5 4b 학습한 대로")
2
+ = Qwen3.5-4B chat_template.jinja 를 그대로 옮기고 KORMo 토큰만 바꿈:
3
+ <|im_start|> → <|BOT|> · <|im_end|> → <|EOT|> · 이미지/영상 <|vision_start|><|image_pad|><|vision_end|> → <|image|> / <|video|>
4
+ KORMo 에서 가져온 것: 이미지 바로 뒤 글 앞에 줄바꿈 하나 (KORMo-VL 원래 방식 · 1.5m 까지 학습한 JSON 판 render 와 같음)
5
+ 그대로인 것(Qwen3.5): tools 를 system 맨 앞에 · XML 인자 tool_call (<function=…><parameter=…>) · 생각 규칙 · <tool_response> · 생성 시작부(기본 생각 켬)
6
+ <tool_call>·<tool_response>·<think> 는 KORMo 단어장에도 한 토큰. <function= · <parameter= 는 두 모델 모두 일반 글. -#}
7
+ {%- set image_count = namespace(value=0) %}
8
+ {%- set video_count = namespace(value=0) %}
9
+ {%- set media_prev = namespace(value=false) %}
10
+ {%- macro render_content(content, do_vision_count, is_system_content=false) %}
11
+ {%- set media_prev.value = false %}
12
+ {%- if content is string %}
13
+ {{- content }}
14
+ {%- elif content is iterable and content is not mapping %}
15
+ {%- for item in content %}
16
+ {%- if 'image' in item or 'image_url' in item or item.type == 'image' %}
17
+ {%- if is_system_content %}
18
+ {{- raise_exception('System message cannot contain images.') }}
19
+ {%- endif %}
20
+ {%- if do_vision_count %}
21
+ {%- set image_count.value = image_count.value + 1 %}
22
+ {%- endif %}
23
+ {%- if add_vision_id %}
24
+ {{- 'Picture ' ~ image_count.value ~ ': ' }}
25
+ {%- endif %}
26
+ {{- '<|image|>' }}{%- set media_prev.value = true %}
27
+ {%- elif 'video' in item or item.type == 'video' %}
28
+ {%- if is_system_content %}
29
+ {{- raise_exception('System message cannot contain videos.') }}
30
+ {%- endif %}
31
+ {%- if do_vision_count %}
32
+ {%- set video_count.value = video_count.value + 1 %}
33
+ {%- endif %}
34
+ {%- if add_vision_id %}
35
+ {{- 'Video ' ~ video_count.value ~ ': ' }}
36
+ {%- endif %}
37
+ {{- '<|video|>' }}{%- set media_prev.value = true %}
38
+ {%- elif 'text' in item %}
39
+ {{- ('\n' if media_prev.value else '') + item.text }}{%- set media_prev.value = false %}
40
+ {%- else %}
41
+ {{- raise_exception('Unexpected item type in content.') }}
42
+ {%- endif %}
43
+ {%- endfor %}
44
+ {%- elif content is none or content is undefined %}
45
+ {{- '' }}
46
+ {%- else %}
47
+ {{- raise_exception('Unexpected content type.') }}
48
+ {%- endif %}
49
+ {%- endmacro %}
50
+ {%- if not messages %}
51
+ {{- raise_exception('No messages provided.') }}
52
+ {%- endif %}
53
+ {%- if tools and tools is iterable and tools is not mapping %}
54
+ {{- '<|BOT|>system\n' }}
55
+ {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }}
56
+ {%- for tool in tools %}
57
+ {{- "\n" }}
58
+ {{- tool | tojson }}
59
+ {%- endfor %}
60
+ {{- "\n</tools>" }}
61
+ {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
62
+ {%- if messages[0].role == 'system' %}
63
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
64
+ {%- if content %}
65
+ {{- '\n\n' + content }}
66
+ {%- endif %}
67
+ {%- endif %}
68
+ {{- '<|EOT|>\n' }}
69
+ {%- else %}
70
+ {%- if messages[0].role == 'system' %}
71
+ {%- set content = render_content(messages[0].content, false, true)|trim %}
72
+ {{- '<|BOT|>system\n' + content + '<|EOT|>\n' }}
73
+ {%- endif %}
74
+ {%- endif %}
75
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
76
+ {%- for message in messages[::-1] %}
77
+ {%- set index = (messages|length - 1) - loop.index0 %}
78
+ {%- if ns.multi_step_tool and message.role == "user" %}
79
+ {%- set content = render_content(message.content, false)|trim %}
80
+ {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
81
+ {%- set ns.multi_step_tool = false %}
82
+ {%- set ns.last_query_index = index %}
83
+ {%- endif %}
84
+ {%- endif %}
85
+ {%- endfor %}
86
+ {%- if ns.multi_step_tool %}
87
+ {{- raise_exception('No user query found in messages.') }}
88
+ {%- endif %}
89
+ {%- for message in messages %}
90
+ {%- set content = render_content(message.content, true)|trim %}
91
+ {%- if message.role == "system" %}
92
+ {%- if not loop.first %}
93
+ {{- raise_exception('System message must be at the beginning.') }}
94
+ {%- endif %}
95
+ {%- elif message.role == "user" %}
96
+ {{- '<|BOT|>' + message.role + '\n' + content + '<|EOT|>' + '\n' }}
97
+ {%- elif message.role == "assistant" %}
98
+ {%- set reasoning_content = '' %}
99
+ {%- if message.reasoning_content is string %}
100
+ {%- set reasoning_content = message.reasoning_content %}
101
+ {%- else %}
102
+ {%- if '</think>' in content %}
103
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
104
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
105
+ {%- endif %}
106
+ {%- endif %}
107
+ {%- set reasoning_content = reasoning_content|trim %}
108
+ {%- if loop.index0 > ns.last_query_index %}
109
+ {{- '<|BOT|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }}
110
+ {%- else %}
111
+ {{- '<|BOT|>' + message.role + '\n' + content }}
112
+ {%- endif %}
113
+ {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %}
114
+ {%- for tool_call in message.tool_calls %}
115
+ {%- if tool_call.function is defined %}
116
+ {%- set tool_call = tool_call.function %}
117
+ {%- endif %}
118
+ {%- if loop.first %}
119
+ {%- if content|trim %}
120
+ {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
121
+ {%- else %}
122
+ {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }}
123
+ {%- endif %}
124
+ {%- else %}
125
+ {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }}
126
+ {%- endif %}
127
+ {%- if tool_call.arguments is defined %}
128
+ {%- for args_name, args_value in tool_call.arguments|items %}
129
+ {{- '<parameter=' + args_name + '>\n' }}
130
+ {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %}
131
+ {{- args_value }}
132
+ {{- '\n</parameter>\n' }}
133
+ {%- endfor %}
134
+ {%- endif %}
135
+ {{- '</function>\n</tool_call>' }}
136
+ {%- endfor %}
137
+ {%- endif %}
138
+ {{- '<|EOT|>\n' }}
139
+ {%- elif message.role == "tool" %}
140
+ {%- if loop.previtem and loop.previtem.role != "tool" %}
141
+ {{- '<|BOT|>user' }}
142
+ {%- endif %}
143
+ {{- '\n<tool_response>\n' }}
144
+ {{- content }}
145
+ {{- '\n</tool_response>' }}
146
+ {%- if not loop.last and loop.nextitem.role != "tool" %}
147
+ {{- '<|EOT|>\n' }}
148
+ {%- elif loop.last %}
149
+ {{- '<|EOT|>\n' }}
150
+ {%- endif %}
151
+ {%- else %}
152
+ {{- raise_exception('Unexpected message role.') }}
153
+ {%- endif %}
154
+ {%- endfor %}
155
+ {%- if add_generation_prompt %}
156
+ {{- '<|BOT|>assistant\n' }}
157
+ {%- if enable_thinking is defined and enable_thinking is false %}
158
+ {{- '<think>\n\n</think>\n\n' }}
159
+ {%- else %}
160
+ {{- '<think>\n' }}
161
+ {%- endif %}
162
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,222 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlavaOnevisionForConditionalGeneration"
4
+ ],
5
+ "bos_token_id": 125030,
6
+ "dtype": "bfloat16",
7
+ "eos_token_id": 125040,
8
+ "image_grid_pinpoints": [
9
+ [
10
+ 384,
11
+ 384
12
+ ],
13
+ [
14
+ 384,
15
+ 768
16
+ ],
17
+ [
18
+ 384,
19
+ 1152
20
+ ],
21
+ [
22
+ 384,
23
+ 1536
24
+ ],
25
+ [
26
+ 384,
27
+ 1920
28
+ ],
29
+ [
30
+ 384,
31
+ 2304
32
+ ],
33
+ [
34
+ 768,
35
+ 384
36
+ ],
37
+ [
38
+ 768,
39
+ 768
40
+ ],
41
+ [
42
+ 768,
43
+ 1152
44
+ ],
45
+ [
46
+ 768,
47
+ 1536
48
+ ],
49
+ [
50
+ 768,
51
+ 1920
52
+ ],
53
+ [
54
+ 768,
55
+ 2304
56
+ ],
57
+ [
58
+ 1152,
59
+ 384
60
+ ],
61
+ [
62
+ 1152,
63
+ 768
64
+ ],
65
+ [
66
+ 1152,
67
+ 1152
68
+ ],
69
+ [
70
+ 1152,
71
+ 1536
72
+ ],
73
+ [
74
+ 1152,
75
+ 1920
76
+ ],
77
+ [
78
+ 1152,
79
+ 2304
80
+ ],
81
+ [
82
+ 1536,
83
+ 384
84
+ ],
85
+ [
86
+ 1536,
87
+ 768
88
+ ],
89
+ [
90
+ 1536,
91
+ 1152
92
+ ],
93
+ [
94
+ 1536,
95
+ 1536
96
+ ],
97
+ [
98
+ 1536,
99
+ 1920
100
+ ],
101
+ [
102
+ 1536,
103
+ 2304
104
+ ],
105
+ [
106
+ 1920,
107
+ 384
108
+ ],
109
+ [
110
+ 1920,
111
+ 768
112
+ ],
113
+ [
114
+ 1920,
115
+ 1152
116
+ ],
117
+ [
118
+ 1920,
119
+ 1536
120
+ ],
121
+ [
122
+ 1920,
123
+ 1920
124
+ ],
125
+ [
126
+ 1920,
127
+ 2304
128
+ ],
129
+ [
130
+ 2304,
131
+ 384
132
+ ],
133
+ [
134
+ 2304,
135
+ 768
136
+ ],
137
+ [
138
+ 2304,
139
+ 1152
140
+ ],
141
+ [
142
+ 2304,
143
+ 1536
144
+ ],
145
+ [
146
+ 2304,
147
+ 1920
148
+ ],
149
+ [
150
+ 2304,
151
+ 2304
152
+ ]
153
+ ],
154
+ "image_token_index": 125041,
155
+ "model_type": "llava_onevision",
156
+ "multimodal_projector_bias": true,
157
+ "pad_token_id": 125032,
158
+ "projector_hidden_act": "gelu",
159
+ "text_config": {
160
+ "_name_or_path": "KORMo-Team/KORMo-10B-sft",
161
+ "architectures": [
162
+ "KORMoForCausalLM"
163
+ ],
164
+ "attention_bias": false,
165
+ "attention_dropout": 0.0,
166
+ "auto_map": {
167
+ "AutoConfig": "_configuration_kormo.KORMoConfig",
168
+ "AutoModel": "_modeling_kormo.KORMoModel",
169
+ "AutoModelForCausalLM": "_modeling_kormo.KORMoForCausalLM"
170
+ },
171
+ "bos_token_id": 125030,
172
+ "dtype": "bfloat16",
173
+ "eos_token_id": 125040,
174
+ "head_dim": 128,
175
+ "hidden_act": "silu",
176
+ "hidden_size": 4096,
177
+ "initializer_range": 0.02,
178
+ "intermediate_size": 16384,
179
+ "mask_type": null,
180
+ "max_position_embeddings": 131072,
181
+ "mlp_bias": false,
182
+ "model_type": "kormo",
183
+ "num_attention_heads": 32,
184
+ "num_hidden_layers": 40,
185
+ "num_key_value_heads": 8,
186
+ "pad_token_id": 125032,
187
+ "pretrain_tp": 1,
188
+ "pretraining_tp": 1,
189
+ "rms_norm_eps": 1e-05,
190
+ "rope_parameters": {
191
+ "rope_theta": 8000000.0,
192
+ "rope_type": "default"
193
+ },
194
+ "rope_theta": 8000000.0,
195
+ "tie_word_embeddings": false,
196
+ "tie_word_embeddins": false,
197
+ "use_cache": true,
198
+ "vocab_size": 125184
199
+ },
200
+ "tie_word_embeddings": false,
201
+ "transformers_version": "5.12.1",
202
+ "use_cache": false,
203
+ "video_token_index": 125042,
204
+ "vision_aspect_ratio": "anyres_max_9",
205
+ "vision_config": {
206
+ "attention_dropout": 0.0,
207
+ "dtype": "bfloat16",
208
+ "hidden_act": "gelu_pytorch_tanh",
209
+ "hidden_size": 1152,
210
+ "image_size": 384,
211
+ "intermediate_size": 4304,
212
+ "layer_norm_eps": 1e-06,
213
+ "model_type": "siglip_vision_model",
214
+ "num_attention_heads": 16,
215
+ "num_channels": 3,
216
+ "num_hidden_layers": 27,
217
+ "patch_size": 14,
218
+ "vision_use_head": false
219
+ },
220
+ "vision_feature_layer": -1,
221
+ "vision_feature_select_strategy": "full"
222
+ }
generation_config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 125030,
4
+ "eos_token_id": [
5
+ 125040,
6
+ 125040
7
+ ],
8
+ "pad_token_id": 125032,
9
+ "transformers_version": "5.12.1"
10
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30f5b4027bb16ce26a74bf531da7a962bbc01a086853300671b9c914ec10de92
3
+ size 22382349432
processor_config.json ADDED
@@ -0,0 +1,206 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_processor": {
3
+ "do_convert_rgb": true,
4
+ "do_normalize": true,
5
+ "do_pad": true,
6
+ "do_rescale": true,
7
+ "do_resize": true,
8
+ "image_grid_pinpoints": [
9
+ [
10
+ 384,
11
+ 384
12
+ ],
13
+ [
14
+ 384,
15
+ 768
16
+ ],
17
+ [
18
+ 384,
19
+ 1152
20
+ ],
21
+ [
22
+ 384,
23
+ 1536
24
+ ],
25
+ [
26
+ 384,
27
+ 1920
28
+ ],
29
+ [
30
+ 384,
31
+ 2304
32
+ ],
33
+ [
34
+ 768,
35
+ 384
36
+ ],
37
+ [
38
+ 768,
39
+ 768
40
+ ],
41
+ [
42
+ 768,
43
+ 1152
44
+ ],
45
+ [
46
+ 768,
47
+ 1536
48
+ ],
49
+ [
50
+ 768,
51
+ 1920
52
+ ],
53
+ [
54
+ 768,
55
+ 2304
56
+ ],
57
+ [
58
+ 1152,
59
+ 384
60
+ ],
61
+ [
62
+ 1152,
63
+ 768
64
+ ],
65
+ [
66
+ 1152,
67
+ 1152
68
+ ],
69
+ [
70
+ 1152,
71
+ 1536
72
+ ],
73
+ [
74
+ 1152,
75
+ 1920
76
+ ],
77
+ [
78
+ 1152,
79
+ 2304
80
+ ],
81
+ [
82
+ 1536,
83
+ 384
84
+ ],
85
+ [
86
+ 1536,
87
+ 768
88
+ ],
89
+ [
90
+ 1536,
91
+ 1152
92
+ ],
93
+ [
94
+ 1536,
95
+ 1536
96
+ ],
97
+ [
98
+ 1536,
99
+ 1920
100
+ ],
101
+ [
102
+ 1536,
103
+ 2304
104
+ ],
105
+ [
106
+ 1920,
107
+ 384
108
+ ],
109
+ [
110
+ 1920,
111
+ 768
112
+ ],
113
+ [
114
+ 1920,
115
+ 1152
116
+ ],
117
+ [
118
+ 1920,
119
+ 1536
120
+ ],
121
+ [
122
+ 1920,
123
+ 1920
124
+ ],
125
+ [
126
+ 1920,
127
+ 2304
128
+ ],
129
+ [
130
+ 2304,
131
+ 384
132
+ ],
133
+ [
134
+ 2304,
135
+ 768
136
+ ],
137
+ [
138
+ 2304,
139
+ 1152
140
+ ],
141
+ [
142
+ 2304,
143
+ 1536
144
+ ],
145
+ [
146
+ 2304,
147
+ 1920
148
+ ],
149
+ [
150
+ 2304,
151
+ 2304
152
+ ]
153
+ ],
154
+ "image_mean": [
155
+ 0.5,
156
+ 0.5,
157
+ 0.5
158
+ ],
159
+ "image_processor_type": "LlavaOnevisionImageProcessor",
160
+ "image_std": [
161
+ 0.5,
162
+ 0.5,
163
+ 0.5
164
+ ],
165
+ "resample": 3,
166
+ "rescale_factor": 0.00392156862745098,
167
+ "size": {
168
+ "height": 384,
169
+ "width": 384
170
+ }
171
+ },
172
+ "image_token": "<|image|>",
173
+ "num_image_tokens": 729,
174
+ "processor_class": "LlavaOnevisionProcessor",
175
+ "video_processor": {
176
+ "data_format": "channels_first",
177
+ "default_to_square": false,
178
+ "do_convert_rgb": true,
179
+ "do_normalize": true,
180
+ "do_pad": true,
181
+ "do_rescale": true,
182
+ "do_resize": true,
183
+ "do_sample_frames": false,
184
+ "image_mean": [
185
+ 0.5,
186
+ 0.5,
187
+ 0.5
188
+ ],
189
+ "image_std": [
190
+ 0.5,
191
+ 0.5,
192
+ 0.5
193
+ ],
194
+ "resample": 3,
195
+ "rescale_factor": 0.00392156862745098,
196
+ "return_metadata": false,
197
+ "size": {
198
+ "height": 384,
199
+ "width": 384
200
+ },
201
+ "video_processor_type": "LlavaOnevisionVideoProcessor"
202
+ },
203
+ "video_token": "<|video|>",
204
+ "vision_aspect_ratio": "anyres_max_9",
205
+ "vision_feature_select_strategy": "full"
206
+ }
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "bos_token": "<|BOS|>",
4
+ "clean_up_tokenization_spaces": false,
5
+ "eos_token": "<|EOT|>",
6
+ "is_local": true,
7
+ "local_files_only": true,
8
+ "max_length": null,
9
+ "model_input_names": [
10
+ "input_ids",
11
+ "attention_mask"
12
+ ],
13
+ "model_max_length": 1000000000000000019884624838656,
14
+ "pad_to_multiple_of": null,
15
+ "pad_token": "<|PAD|>",
16
+ "pad_token_type_id": 0,
17
+ "padding_side": "right",
18
+ "processor_class": "LlavaOnevisionProcessor",
19
+ "tokenizer_class": "TokenizersBackend",
20
+ "torch_dtype": "auto"
21
+ }